Research And Teaching With RevBayes

Overview

  • Introduction to graphical modeling
  • How graphical modeling makes phylogenetic estimation more flexible

The Graphical Modeling Framework

  • What is a model?

The Graphical Modeling Framework

  • What is a model?
    • A model uses mathematics to describe a process, or the behavior of a system
    • Important facets of the generating process of the data are written down as parameters

The Graphical Modeling Framework

  • What is a model?

    • A model uses mathematics to describe a process, or the behavior of a system
    • Important facets of the generating process of the data are written down as parameters

  • A graphical model represents the relationships and dependencies between model paramters

A simple case

  • An archer: Let's say the archer is a pretty good archer
good_shots <- rnorm(n= 100, mean= 0, sd = 1)
hist(good_shots, main="Archer arrow shots",  xlab = "Inches from bullseye")

plot of chunk unnamed-chunk-1

A simple case

  • An archer: Let's say the archer is a pretty bad archer
bad_shots <- rnorm(n= 100, mean= 0, sd = 3)
hist(bad_shots, main="Archer arrow shots",  xlab = "Inches from bullseye")

plot of chunk unnamed-chunk-2

A simple case

  • It's easy to describe a model where we are manipulating one parameter
  • But we rarely have such a simple case when we're analysing biological data

A **not so** simple case

  • It's easy to describe a model where we are manipulating one parameter
  • But we rarely have such a simple case when we're analysing biological data
  • We're describing the mechanisms that generated our observed data
    • In the case of a tree, this is the process of molecular or morphological evolution

A **not so** simple case

  • This is where graphical models come in

A **not so** simple case

  • This is where graphical models come in
  • A graphical model represents the joint probability distribution as a graph

A **not so** simple case

  • This is where graphical models come in
  • A graphical model represents the joint probability distribution as a graph
  • Nodes are variables. Edges represent dependencies between nodes.

A **not so** simple case

  • This is where graphical models come in
  • A graphical model represents the joint probability distribution as a graph
  • Nodes are variables. Edges represent dependencies between nodes.

A **not so** simple case

  • This is where graphical models come in
  • A graphical model represents the joint probability distribution as a graph
  • Nodes are variables. Edges represent dependencies between nodes. Image Title

A **not so** simple case

  • This is where graphical models come in
  • A graphical model represents the joint probability distribution as a graph
  • Nodes are variables. Edges represent dependencies between nodes. Image Title

Pr( data | model )

A **not so** simple case

Image Title

Phylogenetical Complexity

  • This is where graphical models come in
  • A graphical model represents the joint probability distribution as a graph
  • Nodes are variables. Edges represent dependencies between nodes. Image Title

Phylogenetical complexity

Image Title

Phylogenetical complexity

Image Title

Phylogenetical complexity

Image Title

Phylogenetical complexity

Image Title

Phylogenetical complexity

Image Title

Who cares about graphical modeling?

I'm going to make the case that this framework enables radical transparency in both research and pedagogy.

Engaging with underlying assumptions

  • Most phylogenetic estimations treat among-site rate variation as Gamma-distributed
    • Not every site in a nucleotide alignment evolves at the same rate
    • Not every character in a morphological dataset evolves at the same rate

Engaging with underlying assumptions

Yang1994

Yang (1994) demonstrated that modeling ASRV as Gamma distributed with 4 Gamma categories is sufficient

Engaging with underlying assumptions

Yang1994

Implies many low- or zero-rate characters

Engaging with underlying assumptions

Ants5

  • We typically don't collect invarient characters!
  • Wagner (2009) and Harisson and Larson (2016) have argued that the Gamma is not appropriate for ASRV in morphological data, and we should use a lognormal

Engaging with underlying assumptions

  • In previous iterations of software, you might have to email a developer and ask them to implement something
  • RevBayes' graphical model framework emphasizes flexibility. We can simply plug in another ASRV distribution.

Engaging with underlying assumptions

  • Maybe there is an effect here
  • Dan will show us some better ways of looking at posterior samples
  • But, the larger point is that flexibility enabled by modularity means that we were able to immediately implement a model we were interested in

Transparency and documentation

It wouldn't be unfair to say there are some things about RevBayes that are more complex than other phylogenetic softwares

Transparency and documentation

tutorials

Transparency and documentation

tutorials

Each one is available as a markdown file on the website repository

Transparency and documentation

You see an analysis you like? Clone it!

Transparency and documentation

Integrate with various other R packages, like Bookdown, to generate a flexible living textbook for your systematics class!

bookdown::render_book("RB_MCMC_Discrete_Morph.Rmd", "bookdown::gitbook")

Transparency and documentation

We also have the RevKernel, and a set of RevNotebooks ready to go for the Pythonistas.